According to GridStackHub.ai's Blackwell Price Index for June 2026 (Week 22), the cheapest NVIDIA B300 GPU rental is $4.48/hr on CoreWeave (1x B300 SXM, 192GB HBM3e, on-demand), with prices ranging from $4.48 to $6.94/GPU/hr across the 3 cloud providers currently tracked. The B300 surged +49.7% week-over-week in late May 2026 — the largest single-week move for any GPU in the GridStackHub database, signaling strong demand for NVIDIA's latest Blackwell stepping. B300 availability is currently limited to independent cloud providers, with hyperscalers expected to add capacity in H2 2026 as NVIDIA scales Blackwell production. GridStackHub tracks all B300 pricing daily — use the GPU Cost Calculator to project your monthly B300 spend against B200, H200, and 50+ other configurations.
Cheapest verified B300 SXM cloud price (CoreWeave, 192GB HBM3e, on-demand) — Blackwell refined: 192GB HBM3e, improved bandwidth vs B200, on-demand availability. The B300 carries the same 192GB VRAM as B200 with incremental Blackwell stepping improvements. Supply is tight; more providers expected in H2 2026.
NVIDIA B300 Cloud Pricing — Live Table (June 2026)
GridStackHub tracks NVIDIA B300 pricing across 3 cloud providers as of Week 22, 2026. The B300 SXM is available on-demand from CoreWeave, Lambda, and RunPod — with the lowest verified price at $4.48/hr from CoreWeave. The +49.7% WoW surge reflects strong demand against constrained supply:
How we track B300 prices: GridStackHub pulls B300 pricing from direct provider API calls, verified pricing pages, and the Blackwell Price Index (Week 22, 2026). VERIFIED = confirmed via live API or pricing page. B300 pricing data is updated daily. Availability is expanding — set a price alert to be notified when new providers list B300 capacity or prices shift. For the full Blackwell stack including B200, GB200, and B100 pricing, visit the Blackwell Price Index.
| Provider | Instance / Config | GPU VRAM | Pricing Type | Price | Status |
|---|---|---|---|---|---|
| CoreWeave | B300 SXM | 192 GB HBM3e | On-demand | $4.48/hr | VERIFIED |
| Lambda | B300 SXM | 192 GB HBM3e | On-demand | $5.29/hr | VERIFIED |
| RunPod | NVIDIA B300 | 192 GB HBM3e | On-demand | $6.94/hr | VERIFIED |
Data sourced from GridStackHub's live pricing database, June 2, 2026 (Week 22). VERIFIED = confirmed via live provider API or pricing page. All prices are on-demand. B300 supply is constrained; verify availability with provider before committing. Prices subject to change — the +49.7% WoW surge reflects market dynamics.
B300 pricing surged +49.7% WoW in late May 2026. This is the sharpest single-week price move in the GridStackHub database — driven by strong AI lab demand against B300's limited initial cloud supply. The surge mirrors early-stage Blackwell availability dynamics (similar to B200's early 2026 ramp). Prices are expected to stabilize and compress as NVIDIA scales B300 production through H2 2026. Set a GPU price alert to track B300 price movements and be notified when new providers list capacity.
NVIDIA B300 Technical Specifications
The B300 is NVIDIA's refined Blackwell architecture GPU — same generation as B200 but with manufacturing improvements and incremental bandwidth and compute enhancements. Key specs:
| Specification | NVIDIA B300 SXM | NVIDIA B200 SXM |
|---|---|---|
| Architecture | Blackwell (refined) | Blackwell |
| GPU Memory | 192 GB HBM3e | 192 GB HBM3e |
| Memory Bandwidth | 8.0+ TB/s (improved) | 8.0 TB/s |
| FP8 Throughput | ~9,500+ TFLOPS | 9,000 TFLOPS |
| BF16 Throughput | ~4,700+ TFLOPS | 4,500 TFLOPS |
| Memory Type | HBM3e (gen 2+) | HBM3e |
| Min Cloud Price (1 GPU) | $4.48/hr (CoreWeave) | $5.29/hr (Lambda) |
| TDP (Power) | ~1,000W | 1,000W |
| Cloud Availability | Growing (3 providers) | Broader (6 providers) |
| 70B model on 1 GPU (BF16) | Yes — with headroom | Yes — with headroom |
The B300 vs B200 distinction matters most for buyers who want the latest Blackwell stepping: incremental improvements in bandwidth and compute efficiency over B200, at a potentially lower price point per GPU-hour (CoreWeave's $4.48/hr vs B200's $5.29/hr floor). As B300 supply expands, expect pricing to compete aggressively with B200.
B300 vs B200 vs H100: Full Comparison
The B300 sits at the top of the Blackwell stack — above B200 and a full generation above H100. Here is how it stacks up across the specs that matter for AI workloads:
| Spec | NVIDIA B300 SXM | NVIDIA B200 SXM | NVIDIA H100 SXM5 |
|---|---|---|---|
| Architecture | Blackwell (refined) | Blackwell | Hopper |
| GPU Memory | 192 GB HBM3e | 192 GB HBM3e | 80 GB HBM3 |
| Memory Bandwidth | 8.0+ TB/s | 8.0 TB/s | 3.35 TB/s |
| FP8 Throughput | ~9,500+ TFLOPS | 9,000 TFLOPS | 3,958 TFLOPS |
| BF16 Throughput | ~4,700+ TFLOPS | 4,500 TFLOPS | 1,979 TFLOPS |
| Min Cloud Price (1 GPU) | $4.48/hr (CoreWeave) | $5.29/hr (Lambda) | ~$1.74/hr (Lambda) |
| Cost per GB VRAM | $0.0233/GB | $0.0276/GB | $0.0218/GB |
| Inference throughput vs H100 | ~3.5–4× faster | ~3–4× faster | Baseline |
| Cloud Availability | Limited (3 providers) | Growing (6 providers) | Broad (15+ providers) |
| Software Maturity | Early (maturing) | Early (maturing) | Mature (CUDA) |
The headline: B300 delivers the same 192GB memory capacity as B200 but at a lower current floor price ($4.48 vs $5.29/hr), making it the more cost-effective Blackwell choice for workloads that don't require the broader provider availability that B200 offers. The tradeoff is availability depth — B200 has 6 tracked providers vs B300's 3, and hyperscalers aren't yet listing B300.
B300 Availability Tracker: In Stock vs Waitlist
B300 availability varies by provider. Here is the current status as of June 2026:
CoreWeave
On-demand B300 SXM available at $4.48/hr. Best current B300 price. HPC-grade networking and enterprise SLAs. Best for production workloads requiring reliability guarantees.
Lambda
On-demand B300 SXM at $5.29/hr. Consistent availability for 1–4 GPU configs. Lambda's infrastructure is well-suited for inference workloads with competitive pricing.
RunPod
On-demand B300 at $6.94/hr. Flexible billing and serverless option. Higher price reflects RunPod's pay-as-you-go model and convenience factor. Consider for bursty workloads.
B300 availability outlook for H2 2026: NVIDIA is ramping Blackwell production aggressively, with B300 as the primary volume Blackwell SKU. Hyperscalers (Google Cloud, AWS, Azure) are expected to add B300 capacity in Q3–Q4 2026. Independent providers like Nebius, Crusoe, and Vultr are also expected to list B300. GridStackHub will track new B300 providers as they list — set a price alert to be notified.
Compare B300 vs B200 vs H100 for your workload
Enter your model size, batch size, and hours per month — see exact monthly cost for B300, B200, H200, H100, and 50+ configurations side by side.
Open GPU Cost Calculator →When to Use B300: Use Case Guide
Choose B300 when:
- You want the latest Blackwell stepping at the lowest current price. CoreWeave's B300 at $4.48/hr is currently cheaper than B200's floor of $5.29/hr — making B300 the more cost-effective Blackwell choice at the single-GPU level.
- You're running inference for 70B+ parameter models. B300's 192GB HBM3e fits Llama 3.1 70B at BF16 with KV cache headroom on a single GPU — no tensor parallelism required, simpler infrastructure, lower latency.
- High-throughput batch inference is your primary workload. B300's improved Blackwell stepping delivers 3.5–4× higher tokens/second than H100 per GPU, making it the most cost-effective option for high-utilization inference deployments.
- You're building for 2026–2027 and want the newest Blackwell SKU. B300 production is ramping through 2026; workloads built here will benefit from supply expansions and price compression as more providers list capacity.
Choose B200 instead when:
- You need broader provider availability. B200 has 6 tracked providers vs B300's 3. If availability and redundancy matter, B200 offers more options for enterprise procurement.
- Hyperscaler SLAs and compliance are required. AWS, Google Cloud, and Azure B200 is expected before B300 at those providers. If you need hyperscaler infrastructure guarantees, B200 is the near-term option.
- You're on a tight budget and B200's price works. At $5.29/hr vs $4.48/hr, B200 is only ~18% more expensive — within the margin of error for most workloads.
Choose H100 instead when:
- Budget is the primary constraint. H100 at $1.74/hr is ~2.5× cheaper than B300. If your utilization is low or your model fits in 80GB, H100's economics are hard to beat.
- You need immediate, proven on-demand availability. H100 is available from 15+ providers with established SLAs and no early-access friction.
- Your software stack needs full CUDA maturity. H100's CUDA ecosystem is battle-tested. B300's Blackwell software is maturing but still in the early phase.
Ask GridStackHub About B300 Pricing
Get answers from live pricing data — compare B300 vs B200 cost, estimate monthly spend, or find the best B300 option for your workload.
Frequently Asked Questions
Track B300 Prices and Set Alerts
B300 pricing surged +49.7% WoW — the sharpest GPU price move in the GridStackHub database. As NVIDIA scales Blackwell production and more providers list B300 capacity, prices will shift. GridStackHub tracks every change:
Get B300 price alerts
We'll notify you when B300 prices drop, new providers list capacity, or the surge stabilizes. Free — no credit card required.
Compare B300 against every alternative for your workload
Set your model size, batch size, and hours per month — see exact monthly cost for B300, B200, H200, H100, AMD MI300X, and 50+ configurations side by side.
Open GPU Cost Calculator →